Comparing Intent Classification in TensorFlow and Oracle Chatbot

I have created sample set of intents with phrases (five phrases per intent, and ten intents). Using this set of data to train and build classification model with TensorFlow and Oracle Chatbot machine learning. Once model is trained, classifying identical sample phrases with both TensorFlow and Oracle Chatbot to compare results. Using Oracle Chatbot with both Linguistic and Machine Learning models.

Summary:

1. Overall TensorFlow model performs better. The main reason for this - I was training TensorFlow model multiple times, until good learning output (minimized learning loss) was produced.

2. Oracle Chatbot doesn't return information about learning loss after training, this makes it hard to decide if training was efficient or no. As consequence - worse classification results, can be related to slightly less efficient training, simply because you don't get information about training efficiency

3. Classification results score: 93% TensorFlow, 87% Oracle Chatbot Linguistic model, 67% Oracle Chatbot Machine Learning. TensorFlow is better, but Oracle Chatbot Linguistic model is very close. Oracle Chatbot Machine Learning model can be improved, see point 2

Results table (click on it, to see maximized):

TensorFlow

List of intents for TensorFlow is provided in JSON file. Same intents are used to train model in Oracle Chatbot:

TensorFlow classification model is created by training 2-layer neural network. Once training is completed, it prints out total loss for the training. This allows to repeat training, until model is produced with optimal loss (as close as possible to 0): 0.00924 in this case:

TensorFlow classification result is good, it failed to classify only one sentence - "How you work?" This sentence is not directly related to any intent, although I should mention Oracle Chatbot Linguistic model is able to classify it. TensorFlow offers correct classification intent as second option, coming very close to correct answer:

Oracle Chatbot

Oracle Chatbot provides UI to enter intents and sample phrases - same set of intents with phrases is used as for TensorFlow:

Oracle Chatbot offers two training models - linguistic and machine learning based.

Once model is trained, there is no feedback about training loss. We can enter phrase and check intent classification result. Below is sample for Linguistic model classification failure - it fails to classify one of the intents, where sentence topic is not perfectly clear, however same intent is classified well by Oracle Chatbot Machine Learning model:

Oracle Chatbot Machine Learning model fails on another intent, where we want to check for hospital (hospital search) to monitor blood pressure. I'm sure if it would be possible to review training quality loss (may be in the next release?), we could decide to re-train model and get results close to TensorFlow. Classification with Oracle Chatbot Machine Learning model:

Socket.IO Integration with Oracle JET

Socket.IO is a JavaScript library for realtime web applications. It comes in two parts - a client-side library that runs in the browser and a server-side library for Node.js. In this post I will walk you through complete integration scenario with Oracle JET.

Here you can see it in action. Send Event button from JET - sends message through Socket.IO to Node.js server side. Message is handled on server side and response is sent back to client (displayed in browser console):

Server side part with Socket.io is implemented in Node.js application and it runs on Express. To create Node.js application (which is just one json file in the beginning), run command:

npm init

To add Express and Socket.io, run commands:

npm install express --save
npm install socket.io --save

To start Node.js application on Express, run command:

npm start

Double check package.json, it should contain references to Express and Socket.IO:

Here is server side code for Socket.IO (I created server.js file manually). When connection is established with the client, message is printed. Method socket.on listens for incoming messages. Method socket.emit transmits message to client. In both cases we can use JSON structure for payload variable. There is cheatsheet for socket.emit - Socket.IO - Emit cheatsheet. Socket.IO server side:

Socket.IO client side can be installed into JET application with NPM. There is separate section in Oracle JET documentation, where you can read step by step instructions about 3-rd party library installation into Oracle JET - Adding Third-Party Tools or Libraries to Your Oracle JET Application. I would recommend manually include Socket.IO dependency entry into package.json in JET:

Then run command to fetch Socket.IO library into JET application node modules. Next continue with instructions from Oracle JET guide and check my sample code:

npm update

To establish socket connection - import Socket.IO into JET module and use io.connect to establish socket connection. Connect to the end point where Express is running with server side Socket.IO listener. Client side is using same socket.on and socket.emit API methods as server side:

Download sample code from my GitHub repository.

ADF on Docker - Java Memory Limit Tuning for JVM

It might look like a challenge to run Java in Docker environment, by default Java is not aware of Docker memory limits. Check this article for example - Java inside docker: What you must know to not FAIL.  I was able to run WebLogic and ADF (Essential WebLogic Tuning to Run on Docker and Avoid OOM) on Docker previously without Java memory issues, using JAVA_OPTIONS=-XX:+UnlockCommercialFeatures -XX:+ResourceManagement -XX:+UseG1GC. However after Docker upgrade to latest version, these settings didn't help anymore. I did't want to hardcode memory setting with -Xmx.

Java started to consume all available memory in Docker and eventually was killed. You can see this from chart below - memory is growing, killed and after restart growing again:

To solve this behaviour, I have applied settings from Java Platform Group, Product Management Blog - Java SE support for Docker CPU and memory limits. I have replaced JAVA_OPTIONS=-XX:+UnlockCommercialFeatures -XX:+ResourceManagement -XX:+UseG1GC set previously with JAVA_OPTIONS=-XX:+UnlockExperimentalVMOptions -XX:+UseCGroupMemoryLimitForHeap -XX:+UseG1GC.

JAVA_OPTIONS=-XX:+UnlockExperimentalVMOptions - XX:+UseCGroupMemoryLimitForHeap -XX:+UseG1GC did the job - JVM stays in Docker memory limits sharp:

This chart shows Java memory behaviour before and after settings were applied. From March 27th - Java memory is a straight line with JAVA_OPTIONS=-XX:+UnlockExperimentalVMOptions - XX:+UseCGroupMemoryLimitForHeap -XX:+UseG1GC:

Oracle JET Offline Persistence Toolkit - Offline Update Handling

Oracle JET Offline Persistence Toolkit supports offline update, create and delete operations. In this post I will describe update use case. Read previous post related to offline toolkit, where I explain how to handle REST pagination, querying and shredding - REST Paging Support by Oracle Offline Persistence in JET.

This gif shows scenario, where we go to offline mode and then changing data in multiple rows. Data update happens offline and each PATCH request is tracked by offline persistence toolkit:

As soon as we go online (Offline checkbox value is changed in Chrome Developer Tools) - requests executed while offline are replayed automatically against backend server:

We should see, how update flow is implemented in JET in this particular case. Once data is changed, we call submitUpdate function. This function in turn calls JET Model API function save. This triggers PATCH call to back-end to update data. If we are offline, JET offline persistence toolkit, transparently records PATCH request to be able to replay it later while online. There are no specific code changes needed by developer to support offline logic during REST call:

Once we go online, listener is invoked and it calls our function synchOfflineChanges. This function triggers request replay to the backend. This means we can control, when requests are replayed. Besides this, we can control each request which failed to be replayed - this is important, when data conflict happens during update in backend:

Online handler is registered with window.addEventListener in the same module, where persistence manager is defined:

Offline Persistence Toolkit 1.1.1 supports extensive logging. You can update to 1.1.1 version by running: npm install @oracle/offline-persistence-toolkit command:

To enable persistence toolkit logger, add persist/impl/logger module to your target module and call logger.option('level', logger.LEVEL_LOG):

Logger prints useful information about offline update, this helps to debug offline functionality:

Download sample application from GitHub repository.

ADF Declarative Component Example

ADF Declarative Component support is popular ADF framework feature, but in this post I would like to explain it from slightly different angle. I will show how to pass ADF binding and Java bean objects into component through properties, in these cases when component must show data from ADF bindings, such approach could offer robustness and simplify component development.

This is component implemented in the sample app - choice list renders data from ADF LOV and button calls Java bean method to print selected LOV item value (retrieved from ADF bindings):

JDeveloper provides wizard to create initial structure for declarative component:

This is ADF declarative component, it is rendered from our own tag. There are two properties. List binding is assigned with LOV binding object instance and bean property with Java bean instance defined in backing bean scope. In this way, we pass objects directly into the component:

LOV binding is defined in target page definition file, where component is consumed:

Bean is defined in the same project, where page which consumes ADF declarative component is created. We need to define component property type to match bean type, for that reason, we must create class interface in the component library and in target project implement it:

Component and main projects can be in the same JDEV application, we can use JDEV working sets to navigate between projects (when running main project, we dont want to run component project, component project is deployed and reused through ADF JAR library):

Bean interface is defined inside component:

Property for list binding is defined with JUCtrlListBinding type, this allows to pass binding instance directly to the component. Same for bean instance, using interface to define bean instance type, which will be assigned from the page, where component is used:

Declarative component is based on combination of ADF Faces components:

Download sample application from GitHub repository.

Find In Cache By Key ADF BC API Method Usage

What if you need to verify - if row with given key exists in fetched rowset? This could be useful while implementing validation logic. ADF BC API method findByKey - will trigger SQL call and fetch row from DB, if row with given key doesn't exist in fetched rowset. Luckily there is ADF BC API method called findInCacheByKey, this method only checks for row in fetched rowset, without going to DB - very convenient in certain situations, when you actually don't want to bring record from DB, if it wasn't fetched.

Imagine table with pagination feature. First ten rows are fetched and exist in the cache:

Now if we call custom method, where findInCacheByKey is invoked twice - you will see different results. First call is using key from fetched rowset - this call will find a row. Second call is using key, which doesn't belong to the fetched rowset - row is not in cache and call will return zero rows:

Download sample app from my GitHub repository.

REST Paging Support by Oracle Offline Persistence in JET

Oracle Offline Persistence query handler - Oracle Rest Query Handler supports pagination for Oracle ADF BC REST service out of the box. Check my previous post to see how querying works through offline persistence toolkit for ADF BC REST service - Shredding and Querying with Oracle Offline Persistence in JET.

Pagination is a must for large REST resources, its great that Oracle offline persistence toolkit supports it. Let's see it in action.

I navigate through the data with left/right arrows, this triggers REST call with pagination parameters - limit and offset. These are standard parameters supported by ADF BC REST. Requests are executed online:

All pages of data are cached by offline toolkit, if while offline we try to access previously cached page by executing REST request with paging parameters - we will get data from offline toolkit. Now I switch offline and try to navigate to the one of cached pages - data is retrieved from cache automatically:

If I navigate to the page, which was not cached (meaning - not accessed while online) - no results returned. In such situation I can navigate back (paging parameters will be updated) and cached data will be displayed for the page which was cached:

Paging navigation control buttons are calling JS functions to update startIndex:

Sample application is using JET Collection API to execute fetch requests. Collection is extended with getURL function which sets limit and offset parameters to execute paging request:

Once again, make sure to use Oracle Rest Query Handler in offline persistence toolkit configuration:

Fetch function is called through JET Collection API. Start index value is calculated dynamically - this allows to execute paging requests. Same function works online and offline, no need to worry about connection status, all online/offline logic is handled by persistence toolkit:

Sample application for this post is available on GitHub.